There Is No Preview Available For This Item
This item does not appear to have any files that can be experienced on Archive.org.
Please download files in this item to interact with them on your computer.
Show all files
Augmented Reality requires accurate camera pose estimates to correctly render virtual graphics into the real world. In unknown or partially known enviroments, visual SLAM (Simultanesous Localisation and Mapping) can provide this while also estimating the geometry of the enviroment.
This talk argues that the traditional approach to visual SLAM - to jointly optimise camera pose and a 3D map of features at 30Hz - is inappropriate for AR. Instead I propose to build the map from a small sub-set of the observed frames, and show that this approach improves the quality of the generated maps. Further I propose to de-couple the map-building process from frame-to-frame tracking: This allows the use of a wealth of robust tracking techniques, resulting in a SLAM system capable of withstanding some (but not all) the rigors of hand-held AR.